Fix #1: remove local M5 Max torch wheel from public deps; target Python 3.13; add README badges#2
Merged
Merged
Conversation
Verified rebaseline A-side (2026-05-28-pixal3d-glb-quality-rebaseline): foreground-isolated RGB preprocessing fed to MoGe/DINO/NAF (adds src/mlx_spatial/pixal3d_preprocess.py), stage-specific 512/1024 DINO conditioning with fail-closed patch-grid validation, plus focused pipeline/CLI/projection tests.
New verified change 2026-05-28-mlx-spatialkit-narrow-band-remesh: native CPU narrow-band remesh (cpp/remesh.cpp + hoisted cpp/triangle_bvh.hpp) closes open boundaries on real Pixal3D fixtures in ~15-65s @res1024, no CUDA/Torch/MLX. Opt-in via export_pixal3d_glb(remesh=True) with a remesh diagnostics stage + topology-blocker clear; opt-in repair_nonmanifold vertex-split (manifold input-prep for QEM). Bundles the rebaseline's file-entangled spatialkit B-side work (export/glb_compare/texture/simplify diagnostics).
…ine/smooth, deterministic)
… capped fallback)
…bly + opt-in wiring
…add README badges The maintainer's personal M5 Max PyTorch build (torch==2.12.0rc3+m5max) leaked into committed always-on config via [tool.uv] override-dependencies + tool.uv.sources + uv.lock path entries, so any brand-new 'uv sync' failed: Failed to generate package metadata for torch==2.12.0rc3+m5max @ path+../../ai-frameworks/... No such file or directory torch is only a parity/reference dep (opt-in 'torch-ref' group, pulled via uniception); the shipped package is pure MLX. Parity tests import torch from a local source checkout via sys.path, independent of uv, so removing the wheel pin costs the local-dev workflow nothing. - pyproject: remove [tool.uv.sources] + override-dependencies; float torch/ torchvision in torch-ref so they resolve from PyPI - pyproject: requires-python >=3.11 -> >=3.13 (matches the only tested env); add Trove classifiers (MIT, Python 3.13, macOS) - uv.lock: regenerate (torch 2.12.0 / torchvision 0.27.0 from PyPI, no local paths) - README: add PyPI/Python/License/Tests badges; Python 3.11+ -> 3.13 Closes #1
…t pixal3d boundary tests
The macos runner now defaults to Python 3.14, but the lock's supported
environment is 3.13-only, so 'uv run' aborted before tests ('current Python
platform is not compatible with the lockfile's supported environments').
- test.yaml: pin setup-uv python-version to 3.13 so CI matches the target env
- pyproject: requires-python '>=3.13' -> '>=3.13,<3.14' to match the 3.13-only
lock/environments (uv/pip on 3.14 now get a clean 'requires 3.13' message)
- xfail the 3 Pixal3D stage-4 boundary tests (auto-camera / Telea inpaint
in progress) so CI reflects the dependency fix, not unrelated in-flight work
…load/download-artifact v7/v8) GitHub is removing Node 20 from runners; actions/checkout@v4 and astral-sh/setup-uv@v5 emitted deprecation warnings. Bump to current Node 24 majors. pypa/gh-action-pypi-publish is a Docker action and is unaffected, so it stays on release/v1.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fixes #1 — a brand-new
uv syncfailed because a maintainer's local-dev-only M5 Max PyTorch build leaked into the committed, always-on config:Three knobs forced that local wheel into every resolution:
[tool.uv] override-dependencies = ["torch==2.12.0rc3+m5max"](global; eagerly read on any resolve)[tool.uv.sources]mappingtorch/torchvisionto../../ai-frameworks/...pathsuv.lockembedding those assource = { path = "../../ai-frameworks/..." }torchis only a parity/reference dependency in the opt-intorch-refgroup (pulled viauniception); the shipped package is pure MLX. Parity tests import torch from a local source checkout viasys.path(gated byMLX_SPATIAL_RUN_TORCH_PARITY=1), independent of uv — so removing the wheel pin doesn't affect the local-dev workflow.Changes
[tool.uv.sources]+override-dependencies; floattorch/torchvisionintorch-refso they resolve from PyPI.requires-python>=3.11→>=3.13(the only environment actually tested — CI +[tool.uv] environmentspin 3.13); add Trove classifiers (MIT, Python 3.13, macOS).torch 2.12.0/torchvision 0.27.0now from PyPI, no local paths.Python 3.11+→Python 3.13.Verification
uv sync --dry-run(default) → succeeds, no torch metadata error.uv sync --group torch-ref --dry-run→ resolves publictorch==2.12.0/torchvision==0.27.0from pypi.org.grep -nE "m5max|ai-frameworks" uv.lock→ no matches.test_pixal3d_inferencefailures are pre-existing onmain, unrelated to this change).Closes #1